Asterfusion Arm-based DPU SmartNIC solution

The popularity of the DPU is rapidly increasing nowadays. However, do you know how to decide whether your organization requires a Data Processing Unit in yours data center or understand its use cases ? Let’s learn about Asterfusion’s DPU solutions.

With the rapid development of cloud network, artificial intelligence, big data, 5G edge computing, data traffic is growing exponentially.

The growth of CPU computing power couldn’t catch up with the growth of network transmission rate, the demand of specialized computing on network side is highly increasing, thus the SmartNIC card is born to offload various features like virtual network, security, storage, etc.

Offloading services that are not suitable for CPU but processing through DPU( SmartNIC ) is to meet acceleration requirements of the data plane forwarding and reduce the burden of CPU computing power.

Question 1 : Why we need DPU？

In fact, the essence of SmartNIC Card offloading acceleration is heterogeneous computing which is facing challenges in following aspects:

  • The accelerating part accounts for a limited proportion. For example, if the accelerating proportion is 80%, the acceleration no more than 5 times max.
  • Data moving back and forth between the CPU and the accelerators, resulting in an extra consumption in acceleration rates
  • CPU+xPU architecture — -CPU as the center, the entire IO path is too long, and IO becomes the bottleneck of performance

So here Comes DPU (Data Processing Unit)

DPU is a new class of programmable processor, it provides a computing engine for high bandwidth, low latency, and data-intensive computing scenarios, with CPUs and GPUs as one of the three pillars of computing.

The DPU being specialized to operate efficiently on networking packets, storage requests or analytics requests while the CPU is not good at, meanwhile considering the needs of transmission and computing, responsible for traffic scheduling, I/O acceleration, etc.

DPU is essentially a computing engine, can be designed to a SmartNIC or standalone network appliance.
With the born of DPUs, the data center network extends inside the host, DPU realizes traffic scheduling, I/O routing, and acceleration, thus becoming a TRUE fabric endpoint.

Remember True Fabric of Fungible? That’s it!

Questions2: What are the characteristics of DPU?

DPUs need the following key features:

-Well, does Asterfusion sell DPU？
-Of course , remember our SmartNIC card？

-Hold on a minute……we are talking about DPU, what’s the relations between them?
-Just like what we have mentioned above: DPU can be designed to be a SmartNIC or standalone network appliance.

So, Let’s see our SmartNIC characters

Is it a perfect match for DPU’s rich character? If it’s not enough, look at our DPU architecture.

We also provide an open software development environment to help customers develop their own applications with SmartNIC.

Questions 3: What products do we have?

Now, Asterfusion has following DPU solutions:

One is SmartNICs based on Marvel Octeon CN96XX chip which has 2 models. One provides PCIe x16 Gen4.0 lane 4*25G interface and the other provides 2*100G interface.

It enables users to build high-performance intelligent programmable networks while preserving valuable computing resources in servers and reducing the total CAPEX of cloud data centers.

Another is ET series Marvel SOC ARM64 DPU appliance

Based on the Marvell OCTEON CN9132SoC, ET1600 ARM64 DPU appliance can be used as a low cost &power consumption Marvell arm server, also offers security and visibility features like hardware firewall, VPN, Probe and networking features for applications such as IoT/Fog gateways, Edge gateways and much more.

Based on Marvell octeon cn96xx chip，Asterfusion ET3000A ARM64 DPU appliance can be deployed into a variety of network scenarios, such as edge computing , NFV offload, Marvell arm server, also offers security and visibility features like firewall, VPN, Probe and networking features for applications such as Edge gateways and much more.

PS: Standard Linux operating system with DPDK VPP software packages will be provided.

Question 4: What kind of scenarios these DPUs can be used?

  • OVS/VTEP
  • SR-IOV /vhost-user
  • TCP Offload
  • Tunnel encapsulation/unencapsulation
  • 5G UPF offload
  • vFW/vIPS/vLB/vNAT
  • IPsec/SSL
  • XDP/eBPF
  • DPI
  • Compression & decompression
  • NVMe Initiator

If you need it in your network, please do not hesitate. Contact us!

评论

发表评论

此博客中的热门博文

What is AI Data Center Network?

图片
Imagine a world where artificial intelligence doesn’t just analyze data, but creates it. Where AI systems generate stunning works of art, craft compelling stories, and even compose music. This isn’t the stuff of science fiction anymore-it’s the reality of generative AI. And it’s transforming the way we think about data centers. The rise of OpenAI’s ChatGPT and Google’s Bard is just the beginning. The generative AI market is projected to skyrocket to a staggering $126.5 billion by 2031, growing at a mind-boggling 32% per year. At the heart of this revolution are the data centers that power these AI powerhouses. But they’re not your average data centers. They’re specialized AI data center networks, designed from the ground up to meet the unique demands of generative AI. What is AI Data Center Network? AI Data Center Network  refers to a data center network fabric that supports artificial intelligence (AI). It meets the stringent network scalability, performance, and low latency requi...
阅读全文

Difference of LAG,LACP and MLAG

图片
Regardless of whether it's a conventional data center or a modern cloud-based data center, the universal necessity is high availability, and a prevalent issue is the potential single point of failure within the network. In traditional physical networking, to establish highly dependable connections between various devices, LAG (Link Aggregation) technology was developed to ensure high availability of links between servers and switches or even between multiple switches. However, LAG technology possesses its own flaws and is unable to eliminate the risk of equipment failure entirely. Thus, a more advanced technology is required to broaden the original one-to-one connection between devices into a one-to-many connection while ensuring multi-device redundancy and multi-link redundancy. It must also guarantee that devices are able to interconnect with optimal robustness. Introducing MC-LAG, an enhancement of the traditional LAG technology. To accurately comprehend the workings of MC-LAG, ...
阅读全文

Asterfusion Ultra -low latency switch- Ceph cluster deployment and OpenStack integration

图片
  Ceph is a very popular and widely used SDS ( Software Defined Storage ) solution, which can provide object, block and file-level storage at the same time. This article will show you how to use Asterfusion CX-N series ultra-low-latency switches to form a network to carry Ceph storage clusters and use Ceph as the storage backend to integrate with Openstack. Before the popularization of cloud technology, common storage solutions in the IT industry were DAS, SAN, and NAS.  Currently, the three traditional storage solutions have some problems in terms of price, performance, scalability, deployment, operation and maintenance in their respective application scenarios.  So, there is no one-size-fits-all solution on storage. In actual production environments, we often need to make trade-offs based on data size, performance requirements, budget, and application scenarios. The emergence of Ceph is a compromise solution to the above problems. What is Ceph ? Ceph is a very popular o...
阅读全文